Because providing always accurate LLM-based paraphrasing would be much more difficult.